DC25_SLIDES_14_LargeLanguageModelEnabledEngineering_Kumar.pdf

1、Information Classification:GeneralWelcome to ConferenceJanuary 2830,2025Santa Clara Convention Center1ExpoJanuary 2930,2025 Information Classification:GeneralLarge Language Model Enabled Engineering Code Generation Using Novel Data Processing and Augmentation FlowAkhilesh Kumar,(Ansys)Norman Chang,(

2、Ansys)Yu-Chen Lin(Ansys),Wenliang Zhang(Ansys),Muhammad Zakir(Ansys)Rucha Apte(Nvidia),Haiyang He(Ansys),Jyh-Shing Roger Jang(NTU)2Information Classification:General Introduction Challenges LLM Based Code Generation SeaScape Code Generation Framework Novel Data Preprocessing for Code Generation Eval

3、uations and Results ConclusionsAgenda3Information Classification:GeneralSeaScape Platform4Power Noise and Reliability analyses framework for SoC/3DICMain components of SeaScape Infra are Database,Scheduler,Python Interface and GUIElastic Compute and big data analyticsML SupportInformation Classifica

4、tion:GeneralMapReduce on SeaScape5The input data is split into smaller chunksOn each of the smaller data chunks an operation Map is applied.This is parallelized.The above operation provides intermediate results.The intermediate results are combined and is called as Reduce operation.A final result is

5、 generated as the last step.Information Classification:GeneralMapReduce Code in SeaScape6RedHawk-SC supports a custom implementation of MapReduce programming model Supports distributed file systems,scheduler for job parallelization,and communication between the computing resourcesThe RHSC Mapreduce

6、scripts can quickly become more complex for a variety of tasks and queriesEnd users not proficient in Python and RHSC MapReduce programming paradigm find developing such scripts challenging LLMs can significantly improve the productivity and user experience for assisting in developing such scripts u